Dynamical Movement Primitives: Learning Attractor Models for Motor Behaviors

By Auke Jan Ijspeert, Jun Nakanishi, Heiko Hoffmann, Peter Pastor, Stefan Schaal. 2013

1,715 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Statistical and Nonlinear Physics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistical and Nonlinear Physics

Citations in the corpus

Benchmark for Skill Learning from Demonstration: Impact of User Experience, Task Complexity, and Start Configuration on Performance (2019, background): “Based on the choice of motion generation strategy, the majority of skill learning approaches can be further broadly categorized into four groups: time-invariant (stable) dynamical systems [2], [3], [13], time-dependent dynamical systems [7], [9], [10], time-dependent statistical methods [8], [14], [”
An Algorithmic Perspective on Imitation Learning (2018, methodology): “Dynamic movement primitives (DMPs) [Schaal et al., 2004, Ijspeert et al., 2013] and the stable estimator of dynamical systems (SEDS) approach [Khansari-Zadeh and Billard, 2011] are representations that explicitly satisfy the condition of smooth convergence to the goal state.”
Language-Conditioned Imitation Learning for Robot Manipulation Tasks (2020, methodology): “We used motor primitives inspired by the approach in [16].”
Lessons from the Amazon Picking Challenge: Four Aspects of Building Robotic Systems (2016, background): “movement primitives, which represent motion as a set of dynamical systems [23] with few parameters to learn [46], and thus provide suitable assumptions to restrict the space of robot motion.”
Robot Learning System Based on Adaptive Neural Control and Dynamic Movement Primitives (2019, background): “In contrast, the dynamic movement primitive (DMP), which is based on a nonlinear DS [6], only requires one demonstration to model motion; here, the DMP models the movement trajectory as a spring-damper system integrated with an unknown function to be learned.”
Learning Navigation Subroutines from Egocentric Videos (2019, neutral): “Dynamic Motion Primitives (DMPs) propose a framework for specifying macro actions [9], which can be learned from a demonstration [10] for a speciﬁc macro action.”
Learning bimanual end-effector poses from demonstrations using task-parameterized dynamical systems (2015, methodology): “[8] extend dynamic movement primitives (DMPs) [4] with a method that modulates their velocity using a force term computed based on a virtual spring connecting the end-effectors.”
A tutorial on task-parameterized movement learning and retrieval (2015, methodology): “Approaches employing a single model whose parameters are modulated by task parameters, see e.g. [102, 51, 43, 79, 71, 70].”
Learning Partially Contracting Dynamical Systems from Demonstrations (2017, methodology): “Indeed, as noted in [2], a more careful choice of coordinate system would help circumvent the issue of excess curvature during generalization.”
Recent Advances in Robot Learning from Demonstration (2020, methodology): “Several approaches to trajectory learning, such as dynamical systems (70, 83, 97, 98, 100, 101) and probabilistic inference (13, 62, 82), have been studied in the literature.”
Robot Programming without Coding (2020, background): “One of the most commonly used is that based on Dynamic Movement Primitives (DMPs) [5].”
Neural learning of vector fields for encoding stable dynamical systems (2014, methodology): “[13] Auke Ijspeert, Jun Nakanishi, Peter Pastor, Heiko Hoffmann and Stefan Schaal.”
Interaction primitives for human-robot cooperation tasks (2014, background): “A particularly successful approach to imitation learning is based on Dynamic Motor Primitives (DMPs)[6].”
On improving the extrapolation capability of task-parameterized movement models (2013, methodology): “GMR has mostly been used in three manners: 1) as an autonomous system with ξ = [x, ẋ], by learning P(x, ẋ) with a GMM, with x and ẋ representing position and velocity (either in task space or joint space), and by retrieving iteratively velocity commands during reproduction by estimating P(ẋ|x) with ”